Improved Fire Hawk Optimizer with Crossover Scheme for Text Document Clustering

Abstract

Text Document Clustering (TDC) is an important task in document analysis, which groups unstructured text documents based on their similarities. The Fire Hawk Optimizer (FHO) has recently demonstrated strong performance in continuous optimization. However, the original FHO encounters difficulties in maintaining population diversity over time, so getting stuck in the local optimum is likely. This paper proposes an improved version of the FHO algorithm with several strategies to solve the TDC issue. It starts with a guided initialization strategy for enhancing the initial population quality. Furthermore, it integrates multiple crossover operators between the best global solution and any random individual in the population to enhance population diversity and search efficiency without a notable computational overhead. The improved FHO was tested on ten benchmark TDC datasets using four standard clustering metrics: accuracy, F-measure, entropy, and purity. In particular, the improved FHO with one-point crossover achieved average improvements of more than 12% across all metrics, with statistical testing confirming robustness and generalization. The study is one of the first successful applications of FHO to text clustering and demonstrates clear superiority over the state of the art.